Rod Downey Victoria University Wellington New Zealand Madison, June 2010 -...
Transcript of Rod Downey Victoria University Wellington New Zealand Madison, June 2010 -...
Degree Classes via Approximations; and Applications
Rod DowneyVictoria University
WellingtonNew Zealand
Madison, June 2010
THANKS
Support from the University of Chicago.Supported by the Marsden Fund of New Zealand.Support from the FRG Grant.My mother without whom this would have been possible
REFERENCES
Main project joint work with Noam GreenbergTotally ω-computably enumerable degrees and bounding criticaltriples (with Noam Greenberg and Rebecca Weber) Journal ofMathematical Logic, Vol. 7 (2007), 145 - 171.Totally < ωω-computably enumerable degrees and m-toppeddegrees, Proceedings TAMC 2006.Finite Randomness, with Paul Brodhead and Keng meng Ng, inprepWorking with strong reducibilities above totally ω-c.e. degrees,(with George Barmpalias and Noam Greenberg) Transactions ofthe American Mathematical Society, Vol. 362 (2010), 777-813.Some new natural definable degree classes.Hierarchies of definable degrees. (tentative title)
MOTIVATION
Understanding the dynamic nature of constructions, anddefinability in the natural structures of computability theory suchas the computably enumerable sets and degree classes.Beautiful examples: (i) definable solution to Post’s problem ofHarrington and Soare(ii) definability of the double jump classes for c.e. sets of Cholakand Harrington
(iii) (Nies, Shore, Slaman) Any relation on the c.e. degreesinvariant under the double jump is definable in the c.e. degrees iffit is definable in first order arithmetic.The proof of (i) and (ii) come from analysing the way theautomorphism machinery fails. (ii) only gives Lω1,ω definitions.
NATURAL DEFINABILITY
This work is devoted to trying to find “natural” definitions.For instance, the NSS Theorem involves coding a standard modelof arithmetic into the c.e. degrees, using parameters, and thendividing out by a suitable equivalence relation to get the (absolute)definability result.As atriculated by Shore, we seek natural definable classes as perthe following.
A c.e. degree a is promptly simple iff it is not cappable.(Ambos-Spies, Jockusch, Shore, and Soare)
(Downey and Lempp) A c.e. degree a is contiguous iff it is locallydistributive, meaning that
∀a1,a2,b(a1 ∪ a2 = a ∧ b ≤ a→∃b1,b2(b1 ∪ b2 = b∧b1 ≤ a1 ∧ b2 ≤ a2)),
holds in the c.e. degrees.(Ambos-Spies and Fejer) A c.e. degree a is contiguous iff it is notthe top of the non-modular 5 element lattice in the c.e. degrees.
(Downey and Shore) A c.e. truth table degree is low2 iff it has nominimal cover in the c.e. truth table degrees.(Ismukhametov) A c.e. degree is array computable iff it has astrong minimal cover in the degrees.
SECOND MOTIVATION: UNIFICATION
It is quite rare in computability theory to find a single class ofdegrees which capture precisely the underlying dynamics of awide class of apparently similar constructions.Example: promptly simple degrees again.Martin identified the high c.e. degrees as the ones arizing fromdense simple, maximal, hh-simple and other similar kinds of c.e.sets constructions.K-trivials and lots of people, especially Nies and Hirschfeldt.
Our inspiration was the the array computable degrees.These degrees were introduced by Downey, Jockusch and StobThis class was introduced by those authors to explain a number ofnatural “multiple permitting” arguments in computability theory.
Definition: A degree a is called array noncomputable iff for allfunctions f ≤wtt ∅′ there is a a function g computable from a suchthat
∃∞x(g(x) > f (x).
Looks like “non-low2.”Indeed many nonlow2 constructions can be run with only theabove. For example, every anc degree bounds a generic.
c.e. anc degree are those that:(Kummer) Contain c.e. sets of infinitely often maximal Kolmogorovcomplexity(Downey, Jockusch, and Stob) bound disjoint c.e. sets A and Bsuch that every separating set for A and B computes the haltingproblem(Cholak, Coles, Downey, Herrmann) The array noncomputablec.e. degrees form an invariant class for the lattice of Π0
1 classesvia the thin perfect classes
THE FIRST CLASS
(Downey, Greenberg, Weber) We say that a c.e. degree a istotally ω-c.e. iff for all functions g ≤T a, g is ω-c.e.. That is, thereis a computable approximation g(x) = lims g(x , s), and acomputable function h, such that for all x ,
|s : g(x , s) 6= g(x , s + 1)| < h(x).
array computability is a uniform version of this notion where h canbe chosen independent of g.Every array computable degree (and hence every contiguousdegree) is totally ω-c.e..
AND LATTICE EMBEDDINGS
Lattice embedding into the c.e. degrees. (Lerman, Lachlan,Lempp, Solomon etc.)One central notion:(Downey, Weinstein) Three incomparable c.e. degrees a0,b,a1form a weak critical triple iff a0 ∪ b = a1 ∪ b and there is a c.e.degree c ≤ a0,a1 with a0 ≤ b ∪ c.a, b0 and b1 form a critical triple in a lattice L, if a ∪ b0 = a ∪ b1,b0 6≤ a and for d, if d ≤ b0,b1 then d ≤ a.A lattice L has a weak critical triple iff it has a critical triple.
Critical triples attempt to capture the “continuous tracing” neededin an embedding of the lattice M5 below, first embedded byLachlan.
t t t
t
t
@@@@@@@@@
@
@@@
@@
@@@
b1b0a
1
0
FIGURE: The lattice M5
THEOREM (DOWNEY, WEINSTEIN)There are initial segments of the c.e. degrees where no lattice with a(weak) critical triple can be embedded.
THEOREM (DOWNEY AND SHORE)If a is non-low2 then a bounds a copy of M5.
THEOREM (WALK)Constructed a array noncomputable c.e. degree bounding no weakcritical triple,
and hence it was already known that array non-computability wasnot enough for such embeddings.
ANALYSING THE CONSTRUCTION
Track
Gate Ge
HoleHe
Pockets
A1 A2 A 3
ChuteCorralCe
Pe,i : ΦAe 6= Bi (i ∈ 0,1,2,e ∈ ω).
Ne,i,j : Φe(Bi) = Φe(Bj) = f total implies f computable in A,(i , j ∈ 0,1,2, i 6= j ,e ∈ ω.)Associate H〈e,i〉 with Pe,i and gate G<e,i,j> with Ne,i,j .
Balls may be follower balls (which are emitted from holes), or traceballs.x = x i
e,n. that x is a follower is targeted for Ai for the sake ofrequirement Pe,i and is our nth attempt at satisfying Pe,i .
otherwise it is a trace ball: t je,i,m(x) which indicates it is targeted
for Bj and is the mth trace:at any stage s things look like:x i
e,n, tj1e,i,1, t
j2e,i,2, ..., t
jme,i,m
The key observation of Lachlan was that a requirement Ne,i,j isonly concerned with entry of elements into both Bi and Bj betweenexpansionary stages
When a ball is sitting at a hole it either gets released or it gets anew trace.When released a 1-2 sequence, say, moves down together andthen stops at the first unoccupied 1-2 gate. All but the last one areput in the corral. The last one is the lead trace.Here things need to go thru one ball at a time and we retarget thelead trace as a 1-3 sequence or a 2-3 sequence. The current lasttrace is targeted for 1 it is a 1-3 sequence, else a 2-3 sequence.
ONE GATE
The notion of a critical triple is reflected in the behaviour of onegate. This can be made precise with a tree argument.We have a 1-2 sequence with all but the last in the corral.the last needs to get thru. It’s traces while waiting will be a 2-3sequence, say. (or in the case of a critical triple, a sequence with atrace and a trace for the middle set A).once it enters its target set, the the next comes out of corral andso forth.Now suppose that we want to do this below a degree a. We wouldhave a lower gate where thru drop waiting for some permission bythe relevant set D.We know that if d is not totally ω-c.e. then we have a functiong ≤T D, ΓD = g which is not ω-c.e. for any witness f .We force this enumeration to be given in a stage by stage mannerΓD = g[s].We ignore gratuituos changes by the opponent.
Now we try to build a ω approximation to g to force D to givemany permisssions.thus, when the ball and its A-trace drop to the lower gate, then weenumerate an attempt at a ω-c.e. approximation to ΓD(n)[s].This is repeated each time the ball needs some permission.
A CHARACTERIZATION
THEOREM (DOWNEY, GREENBERG, WEBER)(I) Suppose that a is totally ω-c.e.. Then a bounds no weak critical
triple.(II) Suppose that a is not totally ω-c.e.. Then a bounds a weak critical
triple. ‘(III) Hence, being totally ω-c.e. is naturally definable in the c.e.
degrees.
The proof of (i) involves simulating the Downey-Weinsteinconstruction enough and guessing nonuniformly at the ω-c.e.witness.The other direction is a tree argument simutaling the “one gate”scenario, as outlined.
A COROLLARY
Recall, a set B is called superlow if B′ ≡tt ∅′.
THEOREM (DOWNEY, GREENBERG, WEBER)The low degrees and the superlow degrees are not elementarilyequivalent. (Nies question)
Proof: There are low copies of M5.Also: Cor. (DGW) There are c.e. degrees that are totally ω-c.e.and not array computable.
OTHER SIMILAR RESULTS
THEOREM (DOWNEY, GREENBERG, WEBER)A c.e. degree a is totally ω-c.e. iff there are c.e. sets A, B and C ofdegree ≤T a, such that(I) A ≡T B
(II) A 6≤T C(III) For all D ≤wtt A,B, D ≤wtt C.
(Downey and Greenberg) Actually D can be made as the infimum.
PRESENTING REALS
A real A is called left-c.e. if it is a limit of a computablenon-decreasing sequence of rationals.(eg) Ω =
∑U(σ)↓ 2−|σ|, the halting probability.
A c.e. prefix-free set of strings A ∈ 2<ω presents left c.e. real α ifα =
∑σ∈A 2−|σ| = µ(A).
THEOREM (DOWNEY AND LAFORTE)There exist noncomputable left c.e. reals α whose only presentationsare computable.
THEOREM (DOWNEY AND TERWIJN)
The wtt degrees of presentations forms a Σ03 ideal. Any Σ0
3 ideal canbe realized.
THEOREM (DOWNEY AND GREENBERG)The following are equivalent.(I) a is anc
(II) a bounds a left c.e. real α and a c.e. set B <T α such that if Apresents α, then A ≤T B.
A HIERARCHY
Lets re-analyse the 1-3-1 example.With more than one gate then when it drops down, it needs tohave the same consitions met.That is, for each of the f (i) many values j at the first gate there issome value f (j , s) at the second.This suggest ordinal notations.(Strong Notation) Notations in Kleene’s sense, except that we askthat the notation for an ordinal is given by an effective CantorNormal Form.There is no problem for the for ordinals below ε0, and suchnotations are computably unique.
Now we can define for a notation for an ordinal O, a function to beO-c.e. in an analogous was as we did for ω-c.e..e.g. g is 2ω + 3 c.e., if it had a computable approximation g(x , s),which initially would allow at most 3 mind changes.Perhaps at some stage s0, this might change to ω + j for some j ,and hence then we would be allowed j mind changes, and finallythere could be a final change to some j ′ many mind changes.All low2.
ωω
Analysing the 1-3-1 case, you realize that that construction needsat least ωω.
THEOREM (DOWNEY AND GREENBERG)a is not totally < ωω-c.e. iff a bounds a copy of M5.
The proof in one way uses direct simulation of the pinball machineplus “not < ωω” permissions, building functions at the gates. Atgate n build at level ωn for each Pe of higher peiority.In the reverse direction, we use level ω-nonuniform argumentswhere the inductive strategies are based on the failure of theprevious level. Kind of like a level ω version of Lachlannon-diamond, using the Downey-Weinstein construction as abase.Corollary There are c.e. degrees that bound lattices with criticaltriples, yet do not bound copies of M5.
ADMISSIBLE RECURSION
THEOREM (GREENBERG, THESIS)Let α > ω be admissible. Let a be an incomplete α-ce degree. TFAE.(1) a computes a counting of α(2) a bounds a 1-3-1(3) a bounds a critical triple.
Uses a theorem of Shore that if a computes a cofinal sequence iffit computes a counting. Then the weak critical triple machinerycan actually have a limit. (Plus Maass-Freidman)
THEOREM (DOWNEY AND GREENBERG)Let ψ be the sentence “a bounds a critical triple but not a 1-3-1” and letα be admissable. Then α satisfies ψ iff α = ω.
This is the first natural difference between Rω and RωCK1.
Differences in Greenberg’s thesis are all about coding.
m-TOPPED DEGREES
Recall that a c.e. degrees a is called m-topped if it contains a c.e.set A such that for all c.e. W ≤T A, W ≤m A.
THEOREM (DOWNEY AND JOCKUSCH)Incomplete ones exist, and are all low2. None are low.
THEOREM (DOWNEY AND SHORE)If a is a c.e. low2 degree then there is an m-topped incomplete degreeb > a.
THEOREM (DOWNEY AND GREENBERG)Suppose that b is totally < ωω-c.e. Then a bounds no m-toppeddegree.
The point is that making an m-top is kind of like making ∅′ on atree: PhiAe = We implies We ≤m A, with ΦA
e 6= B.(Downey and Greenberg) There is, however, a totally ωω degreethat is an m-top, and arbitarily complex degrees that are not.
EXPLORING THE HIERARCHY
Theorem (Downey and Greenberg) If n 6= m then the classes oftotally ωn-c.e. and totally ωm-degrees are distinct. Also there is ac.e. degree a which is not totally < ωω-c.e. yet is totally ωω-c.e..Also totally < ωω not ωn for any n.This is also true at limit levels higher up.
THEOREM (DOWNEY AND GREENBERG)There are maximal (e.g.) totally ω-c.e. degrees. These are totallyω-c.e. and each degree above is not totally ω-c.e. degree.
Thus they are another definable class.As are maximal totally < ωω-c.e. degrees.
THEOREM (DOWNEY AND GREENBERG)
a is totally ω2-c.e. implies there is some totally ω-c.e. degree b below awith no critical triple embeddable in [b,a].
Question: Are totally ωn-c.e. degrees are all definable.Other assorted results about contiguity higher up.
THE PROMPT CASE
What about zero bottom? It is posssible to get the infimum to bezero.(DG) For the classes C aoove, we can define a notion of beingpromptly C then show that if a is such for the ω case, then itbound a critical triple with infumum 0.etc.
NORMAL NOTATIONS?
THEOREM (DOWNEY AND GREENBERG)Suppose that a is low2. Then there is a notation O relative to which ais totally ω2-c.e.
∆03 nonuniform version of Epstein-Haass-Kramer/Ershov.
FINITE RANDOMNESS
Replace tests by finite tests. Several variations.If no conditions then on ∆0
2 reals MLR and finite random coincide.If the test Vn : n ∈ ω has |Vn| < g(n) for computable g, we say itis computably finitele random. (Ie if it passes all such tests.)
THEOREM (BRODHEAD, DOWNEY, NG)The c.e. degrees a containing no such real are the totally ω-c.e.degrees.
Compare with
THEOREM (DOWNEY AND GREENBERG)The c.e. segrees containing sets of packing dimension 1 are exactlythe anc degrees.
WORKING ABOVE SUCH DEGREES
With George Barmpalias, Noam and I began to look at the effectof being able to compute such a degree, but with strongreducibilities.
THEOREM (BARMPALIAS, DOWNEY, GREENBERG)Every set in (c.e.) a is wtt reducible to a ranked one iff every set in a iswtt reducible to a hypersimple set iff a is totally ω-c.e.
THEOREM (BARMPALIAS, DOWNEY, GREENBERG)A computably enumerable a computes a pair of left c.e. reals with noupper bound in the cL degrees iff a computes a left c.e. real not cLreducible to a random left c.e. real iff a is anc.
CONCLUSIONS
We have defined a new hierarchy of degree classes within low2.This hierarchy unifies many constructions, andProvides new natural degree definable degree classes.Many questions remain. eg, is array computable definable in thedegrees. Are these classes definable in the degrees?Can they be used higher up in relativized form, say?